Enhanced CNN Architecture for Invasive Ductal Carcinoma Detection in Breast Histopathology Images

Invasive Ductal Carcinoma (IDC) is considered the most frequently occurring breast cancer subtypes, and the early detection of IDC is crucial for treatment plan decision and improve therapy outcomes. Therefore, IDC’s automated identification will help the pathologists in diagnosis and provide a valuable second opinion. This study uses a dataset of Breast Histopathology Images, which is publicly accessible on Kaggle for the IDC classification task. This dataset includes 277,524 patches, 198,738 of which are IDC-negative, and 78,786 are IDC-positive images. We trained a novel architecture based on deep convolutional neural networks, and we also trained some predefined deep learning architectures for comparison. The proposed model outperforms the remaining models and achieves 89.5 % accuracy, 89 % F1-score, and the area under the receiver operating characteristic curve for detecting IDC (AUC) is 0.96 on the testing dataset, which is a new state-of-the-art result compared with the latest published approaches of IDC classification.